Scheduling a Computing Node(s)
Jump to navigation
Jump to search
First you must build a queue that describes the configuration of one or more computing nodes for your project. The queue must select nodes from one of the preset queues using either the Torque or PBS scheduler. The preset queues are described in the Introduction named Arrow Computing Queues.
Example of scheduling a Queue using pbs
Assume you want to schedule two of the four GPUs from the a6000 queue using PBS as the scheduler. Further, we want to specify 16 CPUs, 3 GPUs, 150 Gigs of memory, and a maximum running time of 3 hours.
module load pbs/pbs qsub -I -q a6000 -l select=1:ncpus=16:ngpus=2:mem=150g -l walltime=3:0:0
Example of scheduling a second queue using pbs
Assume we have used ssh from a second terminal to start a second programing running on the other two GPUs on the same GPU as above such that all four GPUs will be in use by the two programs.
module load pbs/pbs qsub -I -q a6000 -l select=1:ncpus=16:ngpus=2:mem=150g -l walltime=3:0:0